Another option that I would suggest if the problem still persists is to change bias correction parameters in WhyD_preproc.m
This in certain cases 'dims-down' the whole image there by improving the segmentations. Thanks.